(1) Field of Invention
The present invention relates to a system and method for adaptive memory recall and, more particularly, to a system and method for adaptive memory recall that allows for recall of an original memory of objects when similar objects are encountered in a different context.
(2) Description of Related Art
Traditional memory stores data at a unique address and can recall the data upon presentation of the complete unique address. Traditional associative memories bind information into a single representation. For instance, data is stored based on a particular context, and the data is recalled upon presentation of that context. In contrast, autoassociative memories are capable of retrieving a piece of data upon presentation of only partial information from that piece of data. Humans can do this type of recognition and recall easily. People experience a significant set of percepts (i.e., a perceived element) in one setting, and if they encounter the same set of percepts in a different setting, they recognize them despite the environmental shift. However, machine perception systems will be confused by the context, and unable to recall the original memory.
U.S. Pat. No. 7,333,963, entitled, “Cognitive memory and auto-associative neural network based search engine for computer and network located images and photograph” (hereinafter referred to as the '963 patent) describes an auto-associative neural network that performs cued retrieval for objects in images and character recognition. However, the network described in the '963 patent does not have the facility for becoming confused when familiar objects are found in an unfamiliar context.
In “Neural Networks and Physical Systems with Emergent Collective Computational Abilities” in Proceedings of the National Academy of Sciences, 79(8): 2254-2558, 1982, Hopfield (hereinafter referred to as the Hopfield reference) provides a model of content-addressable memory; however, the model cannot store memories that have a Hamming distance around half the size of another memory. If memories are this similar, they tend to merge.
The Massachusetts Institute of Technology (MIT) Center for Biological and Computational Learning (CBCL) object recognition model is based on the human visual pathway, as described by Chikkerur et al. in “An Integrated Model of Visual Attention Using Shape-Based Features” in Technical Report CBCL paper 278, Massachusetts Institute of Technology, 2009. On static images, the CBCL object recognition model has been shown in extensive comparisons to perform at the level of the best computer vision systems in recognizing objects in the real world, such as cars, pedestrians, buildings, trees, and road, as described by Serre et al. in “Robust Object Recognition with Cortex-Like Mechanisms” in IEEE Transactions on Pattern Analysis and Machine Intelligence”, 29:411-426, 2007. However, Serre et al. apply preprocessing to separate the objects from the background and then recognize objects. This means that CBCL's object recognition model gets confused by context and, in addition, it cannot retrieve the original context of the objects it has stored.
Each of the aforementioned methods exhibit limitations that make them incomplete. Thus, there is a continuing need for an approach for learning to associate significant objects encountered within a particular context with the ability to recall the original memory, even when encountered within a different context.